Here's a curated list of awesome real-time machine learning blogs, videos, tools and platforms, conferences, research papers, etc.
- 🤔 What even is "real-time" Machine Learning?
- 🆚 Traditional ML vs Real-Time ML
- 🛠️ Tools & Workflow Stages
- 🏗️ Real-Time ML Internal Platform Resources
- 🎥 Videos
- 🏢 Vendors / Platforms
- 🎪 Conferences
Real-time Machine Learning (ML) delivers predictions and adapts models with extremely low latency, using fresh, continuously streaming data. It employs online or continual learning to instantly update models with new information, ensuring the most relevant insights for immediate actions. This dynamic approach contrasts with batch processing and is crucial for applications requiring instant responsiveness to changing patterns.
- Real-Time Predictions: Model outputs generated on-demand as data arrives with extremely low latency.
- Real-Time Features: Input attributes derived from real-time, rapidly changing data, processed quickly.
- Real-Time Learning: Continuous model updating (online or continual learning) using new data for adaptation and improvement of model performance over time.
| Aspect | Traditional ML | Real-Time ML |
|---|---|---|
| Data Processing | Processes static, historical datasets in batches. | Continuously ingests and processes streaming data in real-time. |
| Model Training | Models are trained offline using complete datasets. | Models are updated incrementally as new data arrives, often using online learning algorithms. |
| Latency | Can tolerate higher latency in processing and predictions. | Requires low-latency processing and near-instantaneous predictions. |
| Scalability | Typically scales vertically with more powerful hardware. Horizontal scaling is possible with distributed frameworks. | Often requires horizontal scalability to handle high-volume data streams. |
| Infrastructure | Can run on standard computing resources. | Often requires specialized streaming infrastructure like Apache Kafka or Apache Flink. |
| Adaptability | Models are less adaptive to changing patterns without manual retraining. | Models can adapt to concept drift and evolving patterns in real-time. |
| Feature Engineering | Features are often engineered manually and in advance. | Features may be generated on-the-fly or use automated feature extraction techniques. |
| Model Deployment | Models are deployed as static versions, updated periodically. | Models are continuously updated and deployed in a streaming fashion. |
| Use Cases | Effective for predictive analytics, segmentation, and batch or streaming data predictions. | Ideal for fraud detection, real-time bidding, and personalized recommendations. |
| Data Volume | Can work effectively with smaller datasets. | Typically requires larger volumes of data for accurate real-time predictions. |
| Computational Resources | Generally less computationally intensive. | Optimizes computational resource usage by processing data incrementally, reducing the need for reprocessing entire datasets, but may require consistent resource availability for real-time updates. |
| Monitoring | Periodic model performance checks are usually sufficient unless operating in dynamic environments. | Requires continuous monitoring of model performance and data quality. |
| Feedback Loop | Feedback is incorporated in batch updates. | Immediate feedback integration for rapid model adjustments. |
| Complexity | Generally simpler to implement and maintain. | More complex, requiring specialized knowledge in streaming architectures and online learning algorithms. |
| Time-to-Insight | Longer time from data collection to actionable insights. | Near-immediate insights from incoming data streams. |
-
- Volga
- OpenMLDB
- Feathr
- Chronon
- Feathub
- FeatureForm
- Hopsworks
- Kaskada
- Feast
- Caraml
- Butterfree
- Bytehub
- inlined.io
- Tecton
- Fennel
- Chalk
- Zipline
- H2O Feature Store
- Qwak
- Iguazio
- Amazon SageMaker Feature Store
- Vertex AI Feature Store
- Databricks Feature Store
- Snowflake Feature Store
- Microsoft Azure Feature Store
- Volga
-
Picnic
Industry: e-commerce and grocery retail
-
Picnic's Lakeless Data Warehouse This article discusses Picnic's data architecture, including near-real-time data processing for analytics.
-
Running demand forecasting machine learning models at scale This blog post discusses Picnic's implementation of deep learning models for demand forecasting, including real-time prediction challenges.
-
The trade-off between efficiency and being on-time: Optimizing drop times using machine learning This blog post describes Picnic's use of machine learning for real-time optimization of delivery times.
-
-
Netflix
Industry: Media and Entertainment, Streaming Services
-
Supporting Diverse ML Systems at Netflix This article discusses Netflix's Machine Learning Platform (MLP) and how it supports various ML systems, including real-time applications.
-
Optimizing the Netflix Streaming Experience with Data Science This blog post explores how Netflix uses real-time and near real-time algorithms along with machine learning models to optimize streaming experiences.
-
Netflix Recommendations: Beyond the 5 stars (Part 1) This article discusses Netflix's recommendation system, including real-time ranking and machine learning experimentation with online A/B testing.
-
Machine Learning Platform - Netflix Research This research area page describes Netflix's machine learning infrastructure, which supports both real-time high-throughput, low-latency use cases and high-volume batch workflows.
-
Evolution of ML Fact Store This article discusses Axion, Netflix's fact store for ML, which is used for real-time feature logging and offline feature generation to remove training-serving skew.
-
Scaling Media Machine Learning at Netflix This post outlines Netflix's media machine learning infrastructure, including real-time serving and searching of media feature values using systems like Marken.
-
InTune: Reinforcement Learning-based Data Pipeline Optimization for Deep Learning Recommendation Models This research publication discusses Netflix's use of reinforcement learning for optimizing real-time data ingestion in deep learning recommendation models.
-
-
Uber
Industry: Transportation and Technology
-
Real-time Data Infrastructure at Uber This paper describes Uber's real-time data infrastructure, which processes petabytes of data daily to support various use cases including customer incentives, fraud detection, and machine learning model predictions.
-
Michelangelo: Uber's Machine Learning Platform This article introduces Michelangelo, Uber's ML platform that enables internal teams to build, deploy, and operate machine learning solutions at scale, including real-time prediction capabilities.
-
Building Scalable Streaming Pipelines for Near Real-Time Features This blog post discusses how Uber leverages Apache Flink to build real-time streaming pipelines for generating data and features for machine learning models.
-
DeepETA: How Uber Predicts Arrival Times Using Deep Learning This article explains how Uber uses deep neural networks for global ETA prediction, including real-time traffic information and low-latency requirements.
-
Palette Feature Store: Uber's Centralized Feature Management Platform This resource describes Uber's Palette Feature Store, a centralized database of features used across the company for various machine learning projects, supporting both batch and near real-time use cases.
-
Project RADAR: Intelligent Early Fraud Detection RADAR: Uber's AI system for real-time fraud detection and mitigation using time series analysis and pattern mining.
-
How Uber Optimizes the Timing of Push Notifications using ML and Linear Programming Uber's real-time ML system optimizes push notification timing using XGBoost and linear programming for personalized user engagement.
-
Uber's Real-Time Document Check Uber's real-time ML system for instant ID verification, using on-device image quality checks and server-side document processing.
-
Personalized Marketing at Scale: Uber's Out-of-App Recommendation System Uber's real-time ML system for personalized out-of-app recommendations, using location prediction and multi-stage ranking for billions of messages.
-
Stopping Uber Fraudsters Through Risk Challenges Uber's real-time ML system implements risk challenges like penny drop verification to detect and mitigate payment fraud dynamically.
-
-
TikTok
Industry: Social Media, Entertainment, and Technology
-
Monolith: Real-Time Recommendation System With Collisionless Embedding Table This paper describes TikTok's real-time recommendation system, Monolith, which uses collisionless embedding tables and online learning to adapt quickly to changing user preferences.
-
Build TikTok's Personalized Real-Time Recommendation System in Python This tutorial demonstrates how to build a simplified version of TikTok's recommendation system using Python, including a feature store, vector database, and model serving infrastructure.
-
Real-time Data Processing at TikTok This article discusses TikTok's use of technologies like Apache Kafka for real-time data streaming, enabling immediate processing of user interactions and content.
-
The AI Algorithm that got TikTok Users Hooked This blog post explains key components of TikTok's AI algorithm and recommender system, including its self-training AI engine, content tags, and user profiles.
-
-
Meta
Industry: Technology, Social Media, and Artificial Intelligence
-
Spiral: Self-tuning services via real-time machine learning This article introduces Spiral, Meta's system for self-tuning high-performance infrastructure services at scale, using techniques that leverage real-time machine learning.
-
Scaling data ingestion for machine learning training at Meta This blog post discusses Meta's experience building data ingestion and last-mile data preprocessing pipelines responsible for feeding data into AI training models, including real-time and near real-time processing.
-
Meta's approach to machine learning prediction robustness This article outlines Meta's systematic framework for building prediction robustness, including real-time monitoring and auto-mitigation toolsets for calibration robustness.
-
Machine Learning Platform - Meta Research This research area page describes Meta's machine learning infrastructure, which supports both real-time high-throughput, low-latency use cases and high-volume batch workflows.
-
Inside Meta's AI optimization platform for engineers across the company This blog post introduces Looper, Meta's end-to-end AI platform for optimization, personalization, and feedback collection, supporting 700 AI models and generating 4 million AI outputs per second.
-
New AI advancements drive Meta's ads system performance and efficiency This blog post discusses Meta Lattice, a new model architecture that improves ad performance through real-time intent capture and multi-distribution modeling with temporal awareness.
-
AI debugging at Meta with HawkEye This article introduces HawkEye, Meta's toolkit for monitoring, observability, and debuggability of end-to-end machine learning workflows powering ML-based products.
-
How machine learning powers Facebook's News Feed ranking algorithm Facebook's News Feed uses real-time machine learning to personalize content ranking for billions of users, processing thousands of signals to predict engagement and optimize user experience.
-
Scaling the Instagram Explore recommendations system Instagram's Explore uses a multi-stage real-time ML system with Two Tower neural networks and caching to recommend relevant content from billions of options.
-
-
Google
Industry: Technology, Internet Services, and Artificial Intelligence
-
Real-time AI with Google Cloud Vertex AI This blog post introduces Streaming Ingestion for Vertex AI Matching Engine and Feature Store, enabling real-time updates and low-latency retrieval of data for ML models.
-
Streaming analytics solutions | Google Cloud This page describes Google Cloud's streaming analytics solutions for ingesting, processing, and analyzing event streams in real-time.
-
Introduction to Vertex AI Feature Store This documentation explains how Vertex AI Feature Store provides a centralized repository for organizing, storing, and serving ML features in real-time.
-
Real-time Data Infrastructure at Google This paper describes Google's real-time data infrastructure, which processes petabytes of data daily to support various use cases including customer incentives, fraud detection, and machine learning model predictions.
-
-
Spotify
Industry: Music Streaming, Technology, and Entertainment
-
Unleashing ML Innovation at Spotify with Ray This article discusses how Spotify uses Ray to empower ML practitioners, support diverse ML systems, and accelerate the user journey for ML research and prototyping.
-
Product Lessons from ML Home: Spotify's One-Stop Shop for Machine Learning This post details ML Home, Spotify's internal user interface for their Machine Learning Platform, which provides capabilities for tracking experiments, visualizing results, and monitoring deployed models.
-
The Winding Road to Better Machine Learning Infrastructure Through Tensorflow Extended and Kubeflow This blog post outlines Spotify's journey in establishing building blocks for their platformized Machine Learning experience, leveraging TensorFlow Extended (TFX) and Kubeflow.
-
Feature Stores at Spotify: Building & Scaling a Centralized Platform This talk discusses Spotify's approach to building a centralized ML Platform in a highly autonomous organization, focusing on their feature store strategy.
-
Real-time Data Infrastructure at Spotify This paper describes Spotify's real-time data infrastructure, which processes petabytes of data daily to support various use cases including customer incentives, fraud detection, and machine learning model predictions.
-
The Rise (and Lessons Learned) of ML Models to Personalize Content on Home (Part I) Spotify's journey in implementing real-time ML models for personalized content recommendations on their Home page, covering early challenges and system design.
-
The Rise (and Lessons Learned) of ML Models to Personalize Content on Home (Part II) A follow-up on Spotify's ML personalization efforts, focusing on evaluation methods, automated deployment, and scaling challenges.
-
-
Instacart
Industry: E-commerce, Grocery Delivery, Technology
-
Lessons Learned: The Journey to Real-Time Machine Learning at Instacart This article discusses Instacart's transition from batch-oriented ML systems to real-time, including challenges faced and solutions implemented for real-time serving and features.
-
How Instacart Modernized the Prediction of Real Time Availability for Hundreds of Millions of Items This blog post details Instacart's new real-time item availability model, which combines general, trending, and real-time scores to improve prediction accuracy and reduce computation costs.
-
Predicting the real-time availability of 200 million grocery items This article explains how Instacart uses machine learning to predict real-time availability of hundreds of millions of grocery items across the US and Canada, including their optimized scoring pipeline.
-
Griffin: How Instacart's ML Platform Tripled ML Applications in a year This post introduces Instacart's MLOps platform, Griffin, which includes components for real-time recommendations and other ML applications.
-
How Instacart Used Data Streaming to Meet COVID-19 Challenges This case study describes how Instacart leveraged Confluent Cloud for data streaming to create real-time availability models and fraud detection systems.
-
Distributed Machine Learning at Instacart This article discusses Instacart's distributed ML architecture, including parallel fulfillment ML jobs for real-time applications like batching, routing, and ETA prediction.
-
Instacart's Item Availability Architecture: Solving for scale and consistency This article explains the architecture behind Instacart's real-time item availability system, designed for scalability and consistency in grocery stock predictions.
-
Supercharging ML/AI Foundations at Instacart This blog post discusses Instacart's efforts to improve their ML infrastructure, including faster feature onboarding and retrieval, which are crucial for real-time ML applications.
-
Real-time Fraud Detection with Yoda and ClickHouse This article describes how Instacart built Yoda, a real-time fraud detection system powered by ClickHouse, to detect and prevent fraudulent transactions.
-
-
DoorDash
Industry: Food Delivery, Technology, and Logistics
-
Building Scalable Real-Time Event Processing with Kafka and Flink This article discusses DoorDash's migration to a cloud-native streaming platform powered by Apache Kafka and Apache Flink for continuous stream processing and data ingestion into Snowflake.
-
How DoorDash Built an Ensemble Learning Model for Time Series Forecasting This blog post details DoorDash's implementation of ELITE, an ensemble learning model for efficient and accurate time series forecasting, used for weekly order volume predictions and delivery time forecasts.
-
Building a Gigascale ML Feature Store with Redis This article describes how DoorDash optimized their Redis-based feature store to handle tens of millions of reads per second, enabling real-time machine learning predictions at scale.
-
Using ML and Optimization to Solve DoorDash's Dispatch Problem This blog post explains how DoorDash uses machine learning and optimization techniques to solve the complex dispatch problem of efficiently matching orders with drivers.
-
Maintaining Machine Learning Model Accuracy Through Monitoring This article discusses DoorDash's approach to monitoring machine learning models in production, ensuring their continued accuracy and performance.
-
Building Riviera: A Declarative Real-Time Feature Engineering Framework This post describes DoorDash's development of Riviera, a framework for real-time feature engineering that allows data scientists to specify feature computation logic and production requirements through high-level constructs.
-
Engineering Systems for Real-Time Predictions @DoorDash This presentation by Raghav Ramesh discusses DoorDash's approach to structuring machine learning systems in production for robust and wide-scale deployment of real-time predictions.
-
How DoorDash Upgraded a Heuristic with ML to Save Thousands of Canceled Orders DoorDash implemented real-time ML to replace heuristics, reducing order cancellations by predicting and mitigating potential issues.
-
Personalizing the DoorDash Retail Store Page Experience DoorDash uses real-time ML to personalize retail store pages, dynamically adjusting content based on user preferences and behavior.
-
Homepage Recommendation with Exploitation and Exploration DoorDash's homepage employs real-time ML for personalized recommendations, balancing exploitation of known preferences with exploration of new options.
-
Managing Supply and Demand Balance Through Machine Learning DoorDash utilizes real-time ML models to dynamically balance supply and demand, optimizing resource allocation across their platform.
-
3 Changes to Expand DoorDash's Product Search Beyond Delivery DoorDash enhanced their product search with real-time ML, improving relevance and expanding beyond delivery to include pickup and other services.
-
Improving ETAs with multi-task models, deep learning, and probabilistic forecasts DoorDash leverages real-time ML with multi-task models and deep learning to provide more accurate and probabilistic delivery time estimates.
-
Beyond the Click: Elevating DoorDash's personalized notification experience with GNN recommendation DoorDash implements real-time Graph Neural Networks (GNN) to personalize notifications, enhancing user engagement beyond simple click-based recommendations.
-
-
Booking.com
Industry: Travel and Technology
-
Booking.com Launches New AI Trip Planner to Enhance Travel Planning Experience This announcement introduces Booking.com's AI Trip Planner, which uses machine learning models and large language model technology to create a conversational experience for trip planning.
-
Booking.com Enhances Travel Planning with New AI-Powered Features for Easier, Smarter Decisions This article discusses Booking.com's expansion of AI-powered features, including Smart Filter, Property Q&A, and Review Summaries, which use Generative AI to simplify key steps in the trip planning process.
-
Machine Learning in production: the Booking.com approach This article discusses how Booking.com integrates machine learning into every step of the customer journey, detailing their approach to productionizing ML models.
-
Booking.com: Building a Scalable Machine Learning Platform This case study describes how Booking.com built a machine learning platform that scales to support their 200 data scientists and processes 1.5 million nights reserved every day.
-
Booking.com accelerates data analysis with Confluent Cloud This customer story details how Booking.com transitioned from a self-managed Apache Kafka deployment to Confluent Cloud to improve reliability and enhance data management capabilities.
-
Leverage graph technology for real-time Fraud Detection and Prevention
-
-
Grab
Industry: Technology, Ride-Hailing, Food Delivery, and Digital Payments
-
Real-time data ingestion in Grab This article discusses Grab's approach to real-time data ingestion, which enables faster business decisions, optimizes data pipelines, and provides audit trails for fraud detection.
-
How we got more accurate estimated time-of-arrivals in the app while pushing down tech costs This blog post details how Grab consolidated their machine learning models for ETA prediction using their internal ML platform Catwalk, improving accuracy and reducing computing costs.
-
GrabRideGuide, our new AI tool that predicts ride demand hotspots This article introduces GrabRideGuide, a fully automated real-time ride demand finder for Grab driver-partners, which uses AI to analyze past trends and suggest optimal routes.
-
Evolution of Catwalk: Model serving platform at Grab This post discusses Catwalk, Grab's ML model serving platform that powers their real-time decision-making capabilities across various services.
-
Grab App at Scale with ScyllaDB This article describes how Grab uses ScyllaDB for real-time counters to detect fraud, identity, and safety risks, processing billions of events per day.
-
Harnessing AI for public good: Grab's approach to AI Governance This blog post outlines Grab's use of AI and machine learning models for real-time automated decision-making to enhance customer experiences.
-
Unsupervised graph anomaly detection - Catching new fraudulent behaviours This article explores how Grab uses unsupervised graph anomaly detection techniques to identify new fraudulent behaviors in its platform. It covers the modeling approach, challenges, and real-world applications in fraud prevention.
-
-
Didact AI
Industry: Finance, Machine Learning, Stock Trading
- Didact AI: The anatomy of an ML-powered stock picking engine This blog post details the architecture and technology behind Didact AI's machine learning-based stock picking engine, which consistently beat the S&P 500 for over a year on a weekly basis.
-
Glassdoor
Industry: Technology, Job Search and Company Reviews
- Glassdoor uses machine learning to tell users if they're being paid fairly This article discusses how Glassdoor uses machine learning to analyze millions of salary reports and real-time supply and demand trends in local job markets to determine fair pay.
-
Dailymotion
Industry: Video Sharing and Streaming, AdTech
-
Case study Dailymotion - Martech Compass This case study discusses how Dailymotion implemented AI-based personalization to improve its video recommendation system, resulting in a 300% increase in unique video views.
-
AI Video Intelligence: Innovation for Publishers & Broadcasters This blog post introduces Dailymotion's AI video intelligence technology, which analyzes video content to extract insights like themes, sentiment, and contextual markers for improved content tagging and contextual advertising.
-
Dailymotion's Journey to Crafting the Ultimate Content-Driven Video Recommendation Engine with Qdrant Vector Database This article details Dailymotion's implementation of a content-based recommendation system using Qdrant vector database, processing 420 million+ videos and serving 13 million+ recommendations daily.
-
See Real time activity - Dailymotion Help Center This help article explains Dailymotion's Real-time dashboard feature, which allows content creators to analyze views generated on their content or player embeds within the last 60 minutes and 24 hours.
-
How Dailymotion reinvented itself to create a better user experience This article discusses Dailymotion's transformation in 2017, including the development of algorithms to deliver personalized content and the integration of video player technology with a robust adtech platform.
-
-
Coupang
Industry: E-commerce, Technology, and Logistics
-
A symphony of people and technology: An inside look at a Coupang fulfillment center This article details how Coupang uses AI, machine learning, and real-time data in their fulfillment centers to optimize operations and improve employee experiences.
-
How Coupang Conquered South Korean E-commerce This article explains how Coupang uses AI to coordinate tasks for workers and drivers using real-time data, allocating labor and providing optimal routes.
-
Matching duplicate items to improve catalog quality Coupang utilizes real-time deep embedding vectors and FAISS for efficient similarity search to detect duplicate items across millions of products.
-
Overcoming food delivery challenges with data science Coupang Eats employs real-time machine learning for order assignment, dynamic pricing, and ETA predictions in food delivery.
-
-
Slack
Industry: Technology, Communication, and Collaboration Software
-
How Slack sends Millions of Messages in Real Time This blog post details Slack's architecture for sending millions of real-time messages daily across the globe, including their use of Channel Servers, Gateway Servers, and Presence Servers.
-
Slack delivers native and secure generative AI powered by Amazon SageMaker JumpStart This article discusses how Slack implemented generative AI capabilities using Amazon SageMaker JumpStart, ensuring data security and privacy for their customers.
-
Real-Time Messaging Architecture at Slack This article provides insights into Slack's Pub/Sub architecture designed to manage real-time messages at scale, highlighting the challenges of delivering messages across different time zones and regions.
-
How We Built Slack AI To Be Secure and Private This blog post explains Slack's approach to building AI features while maintaining rigorous standards for customer data stewardship, including their principles for secure and private AI implementation.
-
Real Time Messaging API | Node Slack SDK This documentation describes Slack's Real Time Messaging API, which allows developers to receive events and send simple messages to Slack in real-time using WebSocket connections.
-
Privacy principles: search, learning and artificial intelligence This resource outlines Slack's privacy principles for their AI and machine learning implementations, emphasizing their commitment to data privacy and security.
-
Email Classification Slack uses real-time machine learning to classify incoming email addresses for optimal collaboration suggestions.
-
-
Swiggy
Industry: Food Delivery, Technology, and E-commerce
-
Building Rock-Solid ML Systems This blog post explores how Swiggy ensures ML reliability at scale by focusing on best practices that deliver consistent performance across their systems.
-
Swiggy's Generative AI Journey: A Peek Into the Future This article discusses Swiggy's implementation of AI-powered neural search to help users discover food and groceries in a conversational manner, receiving tailored recommendations.
-
Swiggylytics: Swiggy's real-time Analytics SDK This blog post details Swiggy's customizable analytics SDK for real-time data, enabling remote configuration and marking of real-time events.
-
Hyperlocal Forecasting at Scale: The Swiggy Forecasting platform This article discusses Swiggy's centralized forecasting service, which enables end-users to generate accurate forecasts quickly and cost-effectively.
-
Enabling data science at scale at Swiggy: the DSP story This blog post introduces Swiggy's Data Science Platform (DSP), an in-house ML deployment and orchestration platform that supports hundreds of real-time and batch models generating over 1 billion predictions per day.
-
Deploying deep learning models at scale at Swiggy: Tensorflow Serving on DSP This article details Swiggy's implementation of Tensorflow Serving capability on their Data Science Platform for deploying deep learning models at scale.
-
An ML approach for routing payment transactions This blog post explains how Swiggy uses machine learning to optimally route payment transactions to different payment gateways, improving payment success rates.
-
How ML Powers — When is my order coming? Swiggy uses real-time machine learning models to provide accurate delivery time estimates, considering various dynamic factors affecting order fulfillment.
-
-
Nubank
Industry: Financial Technology (Fintech), Digital Banking
-
Presenting Precog, Nubank's Real Time Event AI This article introduces Precog, Nubank's AI solution designed to improve customer service by efficiently routing calls using customer embeddings and features, significantly enhancing the customer experience.
-
Real-time machine learning models in real life This blog post details Nubank's approach to implementing real-time machine learning models, discussing challenges and solutions for fast inference and real-time pipelines.
-
Fklearn: Nubank's machine learning library This article introduces Fklearn, Nubank's in-house machine learning library that powers a broad set of ML models, solving problems from credit scoring to automated customer support chat responses.
-
The potential of sequential modeling in fraud prevention This post discusses how Nubank leverages sequential modeling, an advanced machine learning technique, to detect and prevent fraud in real-time.
-
Nubank acquires Hyperplane to accelerate AI-first strategy This article discusses Nubank's acquisition of Hyperplane, a data intelligence company, to enhance its AI capabilities for providing more personalized financial products and services.
-
Beyond prediction machines Nubank explores causal inference techniques beyond traditional ML for real-time decision-making in credit limits, interest rates, and marketing strategies.
-
-
Replit
Industry: Developer Tools, AI-Powered Coding Platforms
-
Replit launches the first software creation Agent on iOS and Android
This blog highlights Replit Agent, an AI tool that allows users to build, deploy, and host apps in real time using natural language commands, directly from their phones. -
Built SaaS in Just 10 Minutes Using AI with Replit Agent
This article demonstrates how Replit Agent enables real-time development of SaaS applications, including features like inventory tracking and real-time alerts, with minimal user input. -
Collaborative Coding: Real-Time Collaboration on Replit
Replit's collaborative coding feature allows multiple users to work on the same project in real time, similar to Google Docs but for programming. This is ideal for pair programming and educational use cases. -
Ghostwriter AI & Complete Code Beta
Ghostwriter is an AI-powered pair programmer integrated into Replit that provides real-time code suggestions, transformations, and explanations to enhance productivity. -
34x Growth in AI Projects: The Rise of Real-Time Development on Replit
This blog discusses how Replit has become a central platform for real-time AI development, enabling developers to create and deploy applications instantly.
-
-
Noon
Industry: E-commerce
-
Noon.com Case Study - Google Cloud
This case study explains how Noon leverages Google Cloud's infrastructure, including BigQuery and Recommendations AI, to deliver real-time personalized shopping experiences, optimize inventory, and handle traffic spikes during peak events like Yellow Friday. -
Noon Intelligence Software - Real-Time Monitoring and Alerts
This resource highlights Noon’s use of real-time category intelligence, price monitoring, and availability alerts to enhance customer experience and ensure optimal product visibility. -
How Noon Became an E-commerce Giant in UAE
This article explores how Noon uses AI and machine learning to power its recommendation engine, optimize pricing in real time, and improve inventory management for a seamless customer experience. -
Role of Predictive Analytics in Shaping UAE's Retail and E-commerce Future
This blog discusses how Noon employs predictive analytics to enhance its recommendation engine, enabling personalized shopping experiences and improving customer satisfaction.
-
-
Lyft
Industry: Ridesharing, Mobility-as-a-Service
-
Powering Millions of Real-Time Decisions with LyftLearn Serving
This blog explains how Lyft’s platform, LyftLearn Serving, powers hundreds of millions of real-time decisions daily for use cases like price optimization, ETA prediction, and fraud detection, with a focus on managing both data and control planes. -
Building Real-Time Machine Learning Foundations at Lyft
This article discusses Lyft's initiative to integrate streaming data into its ML workflows, enabling real-time anomaly detection, event-driven decisions, and enhanced traffic infrastructure using geohash aggregation. -
ML Feature Serving Infrastructure at Lyft
This post details the architecture of Lyft’s Feature Service, which supports real-time feature availability for online inference with single-digit millisecond latency, serving millions of requests per minute. -
Real-Time ML with Beam at Lyft
Lyft uses Apache Beam to power real-time ML pipelines for critical functions like dynamic pricing, ETA prediction, and traffic-aware routing. The infrastructure processes millions of events per minute with sub-second latency. -
How Lyft Uses AI to Get You Where You Want to Go Faster
This blog highlights how Lyft uses machine learning to provide real-time ETA predictions, dynamic routing based on live traffic data, and personalized destination suggestions based on user behavior. -
How Lyft Stores the Data Powering Their ML Models
This article explains how Lyft ensures low-latency access to feature data for both training and real-time inference by hosting thousands of features in its Feature Serving service. -
ETA (Estimated Time of Arrival) Reliability at Lyft
Lyft leverages real-time ML to enhance ETA reliability by dynamically analyzing driver availability, traffic, and marketplace conditions, with continuous model updates to adapt to changing environments. -
The Recommendation System at Lyft
Lyft’s recommendation system uses real-time ML to dynamically rank ride modes, adapt to marketplace conditions, and personalize user experiences while exploring reinforcement learning for continuous improvement. -
Pricing at Lyft
Lyft’s pricing system leverages real-time ML with online reinforcement learning to dynamically optimize prices, balancing supply and demand while adapting to market conditions. -
How Lyft Predicts a Rider’s Destination for Better In-App Experience
Lyft’s destination prediction system leverages real-time ML and attention mechanisms to dynamically suggest personalized destinations based on historical rides and session context. -
How Lyft Creates Hyper-Accurate Maps from Open-Source Maps and Real-Time Data
Lyft uses real-time ML with GPS data and map-matching algorithms to detect and correct map errors, creating hyper-accurate maps for efficient routing and localization. -
Building Lyft’s Marketing Automation Platform
Lyft’s Symphony platform uses real-time ML and reinforcement learning to optimize marketing decisions, dynamically allocate budgets, and improve campaign performance at scale. -
Fingerprinting Fraudulent Behavior
Lyft uses real-time ML with deep learning architectures to detect fraudulent behavior by analyzing sequential user activity logs and dynamically identifying anomalies. -
From Shallow to Deep Learning in Fraud
Lyft employs real-time ML with deep learning models to detect fraud dynamically, leveraging sequential user behavior and advanced infrastructure for scalable prototype-to-production workflows.
-
-
Wayfair
Industry: E-commerce, Furniture & Home Goods
-
How Vertex AI Helps Wayfair Achieve Real-Time Model Serving
This blog details how Wayfair transitioned from batch inference to real-time model serving using Vertex AI, reducing deployment time from a month to an hour while maintaining safeguards and automation. -
Agent Co-Pilot: Wayfair's Gen-AI Assistant for Digital Sales Agents
Wayfair's Agent Co-Pilot provides live, contextually relevant chat response recommendations to assist sales agents in real time, improving customer service and reducing handle time by 10%. -
How Wayfair Improves Its Feature Engineering with Vertex AI
This article explains how Wayfair uses Vertex AI Feature Store and Pipelines to ensure real-time feature freshness and consistency between training and production environments, enabling low-latency inference. -
Wayfair Uses MLOps with Vertex AI Pipelines to Improve Supply Chain
This blog explores how Wayfair migrated its supply chain ML workflows to Vertex AI Pipelines, enabling real-time data ingestion, model training, evaluation, and inference for better agility and efficiency. -
Wayfair Debuts Muse: AI-Powered Search and Discovery Tool
Muse is an AI-driven tool that uses generative models to provide real-time search and discovery experiences, allowing users to upload images and reimagine their spaces with personalized recommendations. -
Wayfair CTO on Bringing AI Into the Virtual Living Room
This article discusses how Wayfair leverages generative AI for features like Decorify, which provides real-time room visualizations based on user-uploaded images, enhancing the shopping experience. -
Diving Into Wayfair's Machine Learning Odyssey
This podcast highlights Wayfair's journey in adopting machine learning for real-time pricing algorithms, marketing optimization, and generative AI tools like Decorify for personalized shopping experiences. -
Griffin: How Wayfair Leverages Reinforcement Learning to Send Customers Relevant Communications
Wayfair's Griffin leverages real-time ML with reinforcement learning to optimize personalized notification decisions, dynamically adapting to customer behavior and feedback.
-
-
Airbnb
Industry: Travel Services, Hospitality
-
Airbnb Open-Sources its ML Feature Platform Chronon
This blog post discusses Chronon, Airbnb's open-source platform for transforming raw data into machine learning-ready features, enabling real-time feature engineering and operationalization. -
StreamAlert: Real-Time Data Analysis and Alerting
StreamAlert is a serverless framework developed by Airbnb for real-time data analysis and alerting, designed to handle high-throughput data streams efficiently. -
How Airbnb Uses Machine Learning for Smart Dynamic Pricing
This article explains how Airbnb leverages machine learning to implement dynamic pricing, analyzing real-time and historical data to provide optimal pricing recommendations for hosts. -
Production Implementation of Real-Time Intent Classification at Airbnb
This resource details how Airbnb uses its Bighead infrastructure and Deep Thought inference system for real-time classification of guest messages, improving response times and communication efficiency. -
Feature Engineering at Airbnb Using Chronon
An in-depth look at how Airbnb uses Chronon for feature engineering, focusing on transforming raw data into actionable ML features in real time. -
Real-Time Aggregation Features for Machine Learning
This blog explores approaches to solving challenges in using real-time aggregation features for machine learning, relevant to platforms like Airbnb. -
How Airbnb Processes a Million User Events Every Second
This blog explains Airbnb's User Signals Platform, which processes over 1 million user events per second with low latency, enabling real-time insights for recommendations and personalization. -
Dynamic Pricing at Airbnb with Real-Time Adjustments
This article highlights Airbnb’s AI-driven dynamic pricing model that adjusts rental prices in real time based on demand fluctuations and market conditions. -
How Airbnb Uses Machine Learning to Optimize Travel
This resource details how Airbnb personalizes search rankings and optimizes pricing in near real time using machine learning algorithms. -
Building Airbnb Categories with ML and Human-in-the-Loop
Airbnb leverages real-time ML with human-in-the-loop systems to dynamically categorize and rank listings, using continuous feedback and adaptive algorithms to enhance user experience. -
Machine Learning-Powered Search Ranking of Airbnb Experiences
Airbnb's search ranking system for Experiences leverages real-time ML through online scoring infrastructure, enabling dynamic personalization and adaptive ranking based on live user inputs and preferences.
-
-
"Bring the power of machine learning to the world of streaming data"
This video from Google Cloud Next demonstrates how to deploy and manage complete ML pipelines for real-time inference and predictions using Dataflow ML.
Watch here -
"Jukebox: Spotify's Feature Infrastructure"
Explains how Spotify manages features for machine learning, including their approach to feature stores and real-time feature serving.
Watch here -
"Scaling Up Machine Learning in Instacart Search for the 2020 Surge"
Discusses how Instacart scaled up its machine learning capabilities to handle the surge in demand during 2020, likely including real-time aspects of their search system.
Watch here -
"How Booking.com Used Data Streaming to Put Travel Decisions into Customers' Hands"
Explains how Booking.com leveraged data streaming to provide a comprehensive booking experience, including the use of Confluent's data streaming platform.
Watch here -
"Inside Coupang's AI-Powered Fulfillment Center"
Showcases Coupang's newest fulfillment center, highlighting its AI-directed nerve center and army of robots for efficient operations.
Watch here -
"Real-time Machine Learning: Architecture and Challenges"
Explores architectures and challenges in implementing real-time machine learning systems, emphasizing the importance of fresh data and low-latency predictions.
Watch here -
"Batch-scoring vs Real-time ML systems"
Compares batch scoring and real-time machine learning systems, discussing their advantages, disadvantages, and implementation differences.
Watch here -
"Journey to Real-Time ML: A Look at Feature Platforms & Modern RT ML Architectures Using Tecton"
Demonstrates how to build a robust MLOps platform using MLflow and Tecton on Databricks for managing real-time ML models and features, with insights from FanDuel's implementation.
Watch here -
"How to Build a Real Time ML Pipeline for Fraud Prediction"
Demonstrates how to build a machine learning pipeline with real-time feature engineering for fraud detection, using Iguazio's data science platform to streamline the process from data ingestion to model deployment and monitoring.
Watch here -
"Real-Time ML: Features and Inference // Sasha Ovsankin and Rupesh Gupta // MLOps Podcast #135”
Explores challenges and solutions in implementing real-time machine learning features and inference at LinkedIn.
Watch here -
"Need for Speed: Machine Learning in the Era of Real-Time"
Explores the evolution of real-time machine learning, discussing challenges in latency, data freshness, and resource efficiency, while providing insights on implementing RTML solutions.
Watch here -
"Real-Time Event Processing for AI/ML with Numaflow // Sri Harsha Yayi // DE4AI"
Discusses Intuit's development of Numaflow, an open-source platform designed to simplify event processing and inference on streaming data for machine learning applications.
Watch here -
"ML Batch vs streaming vs real-time data processing"
Compares batch, streaming, and real-time data processing for machine learning, discussing misconceptions, costs, and decision-making criteria.
Watch here -
"Machine Learning is Going Real-Time"
Chip Huyen explores the state, use cases, solutions, and challenges of real-time machine learning in production across US and Chinese companies.
Watch here -
"Realtime Stock Market Anomaly Detection using ML Models | An End to End Data Engineering Project"
Demonstrates building a real-time anomaly detection system for stock market data using Quix Streams, Redpanda, and Docker.
Watch here -
"Real Time ML: Challenges and Solutions - Chip Huyen"
Explores challenges in implementing real-time machine learning systems, including latency, train-predict inconsistency, and managing streaming infrastructure, while discussing potential solutions and architectures.
Watch here -
"Real-time ML Model Monitoring with Data Sketches and Apache Pinot"
Demonstrates how Uber leverages Apache Pinot as a data sketch store for ML model monitoring, using data profiling and sketch-based solutions to enable efficient and scalable monitoring across different data sources.
Watch here -
"MLOps vs ML Orchestration // Ketan Umare // MLOps Podcast #183"
Explores real-time machine learning challenges in traffic prediction and fraud detection, highlighting the importance of buffering and damping reactions in ML systems.
Watch here -
"Lessons Learned: The Journey to Real-Time Machine Learning at Instacart"
Guanghua Shu discusses Instacart's transition from batch-oriented to real-time ML systems, covering infrastructure changes, use cases, and key lessons learned in implementing real-time ML for their e-commerce platform.
Watch here -
"Leveraging GraphQL for Continual Learning in Real-Time ML Systems"
Discusses how to set up real-time infrastructure and continual learning using GraphQL for machine learning systems, addressing limitations of batch-training paradigms and enabling adaptive models.
Watch here -
"Real-Time ML Insights with Richmond Alake"
Explores real-time machine learning tools, techniques, and career insights with Richmond Alake, a Machine Learning Architect at Slalom Build, covering his work experiences and AI startups.
Watch here -
"Why real time event streaming pattern is indispensable for an AI native future"
Discusses the importance of distributed event streaming for real-time analytics and AI-powered experiences, exploring its applications in data collection, enrichment, and measuring drift and explainability.
Watch here -
"Realtime Prediction with Machine Learning and Data Transform with Redpanda"
Explores building efficient AI applications using stateless pipelines and WebAssembly-powered streaming data transforms, demonstrating how to simplify data architecture for real-time analytics and machine learning.
Watch here -
"Build Real-time Machine Learning Apps on Generative AI with Kafka Streams"
Stepan Hinger discusses integrating large language models with data streaming using Kafka, demonstrating real-time AI applications for unstructured data analysis, customer support automation, and business intelligence through a framework called Area.
Watch here -
"Feeding ML models with the data from the databases in real-time - DevConf.CZ 2024"
Vojtech Juranek demonstrates how to use Debezium to ingest real-time data from databases into machine learning models, showcasing a live demo with TensorFlow and discussing challenges and solutions in implementing such systems.
Watch here -
"Architecting Data and Machine Learning Platforms"
Marco Tranquillin and Firat Tekiner preview their upcoming book, covering the entire data lifecycle in cloud environments, from ingestion to activation, with a cloud-agnostic approach to data and ML platform architecture.
Watch here -
"Real-Time ML Workflows at Capital One with Disha Singla"
Disha Singla, Senior Director of Machine Learning Engineering at Capital One, discusses democratizing AI through reusable libraries and workflows for citizen data scientists, focusing on time series analysis, anomaly detection, and fraud prevention.
Watch here -
"ML Auto-Retraining: Update Your Model in Real Time"
Discusses real-time ML techniques for cybersecurity, including aggregate engines for adapting to attacker shifts, Redis-based key-value stores for tracking indicators of compromise, and an auto-retraining framework for regularly updating models on different cadences.
Watch here -
"Real-Time Data Processing for ML Feature Engineering | Weiran Liu and Ping Chen"
Discusses Meta's evolution of real-time data processing infrastructure for machine learning, covering applications in recommendation systems, content understanding, and fraud detection, with a focus on their latest platform "Extreme" and its use in real-time feature engineering.
Watch here -
"Building Real-Time ML Pipelines: Challenges and Solutions"
Yaron Haviv discusses challenges in productizing AI/ML, introduces MLOps and feature stores, and demonstrates building real-time ML pipelines using the open-source MLRun framework, with examples of churn prediction and fraud detection use cases.
Watch here -
"Apache Spark and Apache Kafka for Real-Time Machine Learning"
This webinar explores the integration of Apache Kafka and Apache Spark for building scalable real-time machine learning pipelines, covering fundamentals of real-time ML, challenges faced by data teams, and optimal usage of these technologies for data processing and analysis.
Watch here
Vendors that offer end-to-end solutions covering feature engineering, model training, serving, and monitoring.
-
TurboML – A machine learning platform that's reinvented for real-time. All steps in the ML lifecycle - from data ingestion to feature engineering, model training, deployment, and post-deployment monitoring are designed to handle real-time data.
🔹 Real-time Predictions – Get fresh model outputs on-demand with low-latency online inference.
🔹 Real-time Features – Transform recent data streams as live context for your models.
🔹 Continual Learning – Update models dynamically with new data.
🔹 Streaming Integrations – Natively supports real-time data sources.TurboML enables ML teams to iterate quickly by testing hypotheses on live production data. Whether refining ETA predictions based on ride completions or improving fraud detection using chargeback events, TurboML ensures models stay relevant and effective.
-
Databricks – Unified data and AI platform with Delta Live Tables for real-time ML workflows.
-
AWS SageMaker – Fully managed ML service with real-time feature ingestion, training, and model deployment.
-
Google Cloud Vertex AI – A fully managed ML platform that unifies data prep, training, and deployment with AutoML and custom model support for real-time inference.
-
Qwak – An end-to-end AI platform that enables organizations to build, deploy, manage, and monitor machine learning workflows.
-
DataRobot – An AI platform that automates the end-to-end process of building, deploying, and maintaining machine learning models.
-
Abacus.ai – An end-to-end platform for building, deploying, and managing AI models with real-time data and automated monitoring.
-
Dataiku – An end-to-end AI platform that streamlines data preparation, model development, deployment, and governance for enterprises.
-
H2O.ai – An AI and machine learning platform providing tools for building, deploying, and managing models at scale with a focus on automation and high performance.
-
Iguazio – An AI platform that streamlines the deployment and management of machine learning applications, offering tools for pipeline orchestration, model monitoring, and GPU provisioning.
-
Xenonstack – An advanced analytics platform offering end-to-end MLOps, data pipeline management, and AI-driven insights
-
Azure Machine Learning – A comprehensive cloud-based platform for building, deploying, and managing ML models at scale.
-
Modzy – A platform that enables organizations to deploy, connect, and run machine learning models across various environments, including enterprise systems and edge devices, offering fully managed infrastructure, tools, and workflows.
-
ZenML – A framework designed to standardize and streamline machine learning workflows, enabling reproducibility, collaboration, and seamless deployment across diverse environments.
-
Valohai – A machine learning platform for managing and automating ML workflows and deployments.
-
Datatron – An MLOps platform for model management, deployment, and monitoring.
-
ClearML – An MLOps platform designed to streamline the entire machine learning lifecycle.
Focus on feature storage, transformation, and real-time serving.
- Tecton – Real-time feature platform for ML, integrating feature storage, transformation, and serving.
- Hopsworks – Feature store + ML pipeline orchestration for real-time ML.
- Fennel – A real-time feature engineering platform with an efficient CDC-aware engine for fresh, incremental ML computations.
- Chalk AI – A real-time platform for machine learning that enables data teams to declare features and their dependencies using idiomatic Python in online, streaming, and batch environments.
- Featurebyte – A self-service platform that automates feature engineering and deployment for ML models.
Focus on tracking experiments, managing model versions, and monitoring performance.
- Weights & Biases – An AI developer platform that streamlines machine learning workflows, offering tools for experiment tracking, model management, and evaluation of generative AI applications.
- Comet – A full-stack MLOps platform for tracking experiments, managing models, and deploying them to production.
- Galileo – A platform that gives AI teams a way to evaluate, iterate, monitor and protect AI applications at enterprise scale.
Focus on running, deploying, and serving machine learning models in production.
- Seldon – An MLOps platform that enables organizations to deploy, manage, monitor, and explain machine learning models at scale
- Fal.ai – A platform for high-performance AI model inference and training, specializing in generative media with production-ready APIs and serverless deployment.
- Modelbit – A platform for running machine learning models in production.
Focus on integrating AI into business applications and making insights accessible.
- AI Squared – A data and AI integration platform that helps make intelligent insights accessible to all.
- MindsDB – A platform that integrates various artificial intelligence (AI) models with traditional databases or other data management system
| Conference | Date | Location | Format |
|---|---|---|---|
| HumanX | March 9-13, 2025 | Las Vegas | Onsite |
| Current Bengaluru | March 19, 2025 | Bengaluru | Onsite |
| MLConf | March 27, 2025 | New York | Onsite |
| Data Science Salon SEA | April 16, 2025 | Seattle | Hybrid |
| Data Council | April 22-24, 2025 | San Francisco | Onsite |
| Machine Learning Prague 2025 | April 28, 2025 | Prague | Hybrid |
| Smart Data and AI Summit | May 5-6, 2025 | Riyadh | Onsite |
| Data Science Next Conference Europe | May 7-9, 2025 | Amsterdam | Onsite |
| Conference on Machine Learning and Systems (MLSys) 2025 | May 12-15, 2025 | Santa Clara | Onsite |
| Real-Time Analytics Summit 2025 | May 14, 2025 | Virtual | Virtual |
| Current London | May 20-21, 2025 | London | Onsite |
| AI & Big Data Expo North America 2025 | June 4-5, 2025 | Santa Clara | Onsite |
| Data + AI Summit 2025 | June 9-12, 2025 | San Francisco | Hybrid |
| London Tech Week | June 9-13, 2025 | London | Onsite |
| SuperAI | June 18-19, 2025 | Singapore | Onsite |
| RAISE Summit | July 8-9, 2025 | Paris | Onsite |
| The MachineCon by AIM | July 25, 2025 | New York | Onsite |
| Ai4 | August 11-13, 2025 | Las Vegas | Onsite |
Your contributions are always welcome! Please read the contribution guidelines first.
This awesome list is under the MIT License.